CellMatch v1.0.0

CellMatch package for cross-platform or cross-species, reference-query matching

DESCRIPTION

@@ -0,0 +1,20 @@
+Package: cellmatch
+Type: Package
+Title: Interpretable Matching of Cell Type Profiles
+Version: 0.0.0.9000
+Author: Eric Kernfeld
+Maintainer: Eric Kernfeld <[email protected]>
+Description: Pairs cellular profiles from reference and query datasets in
+    a way that handles imperfection and ambiguity through model
+    selection and interpretable visualizations.
+    Designed for analysis of transcriptomic data from stem cells and 
+    embryo atlases.
+License: GNU GPLv3
+Encoding: UTF-8
+LazyData: true
+RoxygenNote: 7.0.2
+Suggests: testthat (>= 2.1.0), knitr, rmarkdown, Seurat
+Imports: magrittr, Matrix, ggplot2, reshape2
+VignetteBuilder: knitr
+NeedsCompilation: no
+Packaged: 2020-03-27 22:13:04 UTC; erickernfeld

NAMESPACE

@@ -0,0 +1,31 @@
+# Generated by roxygen2: do not edit by hand
+
+export(AggregateByCluster)
+export(ClusterGenes)
+export(CountSupportingGenes)
+export(EvaluateByGene)
+export(ExamineRobustness)
+export(GetDefaultDistances)
+export(GetDefaultTransformations)
+export(GetReasonableAlternative)
+export(MergeMatrixList)
+export(MutateModel)
+export(PadMatrix)
+export(PairedHeatmap)
+export(PlotEquivalentsTSNE)
+export(PlotMissingGenes)
+export(PlotNeighboringModels)
+export(RunCellMatch)
+export(SampleModels)
+export(SelectInformativeGenes)
+export(SelectModelFast)
+export(SelectModelThoroughly)
+export(StudyModel)
+export(convert_species_rownames)
+export(get_cpm)
+export(get_ortholog)
+export(get_ortholog_table)
+export(has_ortholog)
+export(merge_dupe_rows)
+import(ggplot2)
+import(magrittr)

R/DiscreteFunctionTools.R

@@ -0,0 +1,122 @@
+#' Minimize the given function by coordinate descent on a finite domain.
+#'
+#' @param FUN Function to minimize. Should take a list or vector as the first input.
+#' @param init Initialization -- first list to put into FUN.
+#' @param verbose If true, print optimization updates.
+#' @param values What to iterate over at each coordinate.
+#' @param ... Additional args passed to FUN.
+#'
+CoordinateDescent = function( FUN, values, init = NULL, niter = 10, verbose = T, ... ){
+  assertthat::assert_that( all( init %in% values ) )
+  if(is.null(init)){
+    stop("You must specify 'init'.\n")
+  }
+
+  x = init
+  best_in = init
+  best_out = FUN(init, ...)
+  objective = matrix(nrow = niter, ncol = length(init), data = NA )
+  for( ii in 1:niter){
+    previous_out = best_out
+    for(coord in seq_along(init)){
+      x = best_in
+      for(val in values){
+        x[[coord]] = val
+        current_out = FUN(x, ...)
+        if(current_out < best_out){
+          best_out = current_out
+          best_in = x
+        }
+      }
+      objective[ii, coord] = best_out
+      if(verbose){
+        best_in_display = best_in
+        best_in_display[coord] %<>% paste0("|", "|")
+        print(best_in_display)
+        print(best_out)
+      }
+    }
+    if(previous_out <= best_out){
+      break
+    }
+  }
+  return(list(
+    x = best_in,
+    objective = objective,
+    objective_final = best_out
+  ))
+}
+
+
+#' At each coordinate of init, omit that coordinate and record the output.
+#'
+#' @param FUN Function to minimize. Should take a list or vector as the first input.
+#' Must also accept an optional argument 'index_omit' giving the coordinate to omit.
+#' The function itself must know how to omit a coordinate in a manner appropriate to the context.
+#'
+#' @param values Ignored. This is only present for consistency with similar functions.
+#' @param init Initialization -- first list to put into FUN.
+#' @param ... Additional args passed to FUN.
+#'
+OmitSingleCoords = function( FUN, init = NULL, values = NULL, ... ){
+  objective = matrix(nrow = 1, ncol = length(init), data = NA )
+  for(coord in seq_along(init)){
+    objective[1, coord] = FUN( init, index_omit = coord, ...)
+  }
+  rownames(objective) = "omitted"
+  colnames(objective) = names(init)
+  return(list( init = init,
+               objective = objective,
+               objective_at_init = FUN(init, ...)))
+}
+
+#' At each coordinate of init, cycle through all possible values and record the output.
+#'
+#' @param FUN Function to minimize. Should take a list or vector as the first input.
+#' @param init perturbations of each coordinate of this input are studied.
+#' @param values What to iterate over at each coordinate.
+#' @param ... Additional args passed to FUN.
+#'
+ComputeSingleCoordPaths = function( FUN, values, init, niter = 10, ... ){
+  assertthat::assert_that( all( init %in% values ) )
+  objective = matrix(nrow = length(values), ncol = length(init), data = NA )
+  for(coord in seq_along(init)){
+    for(ii in seq_along(values)){
+      current_in = init
+      current_in[[coord]] = values[[ii]]
+      objective[ii, coord] = FUN(current_in, ...)
+    }
+  }
+  rownames(objective) = values
+  colnames(objective) = names(init)
+  return(list( init = init,
+               objective = objective))
+}
+
+
+#' Sample inputs uniformly at random.
+#'
+#' @param FUN Function to minimize. Should take a list or vector as the first input.
+#' @param init If coords_fix is given, \code{init[coords_fix]} remains under the null model and only
+#' the remaining entries are sampled.
+#' Otherwise, only the length matters, as it determines the dimension of the cube to be sampled.
+#' @param values What to iterate over at each coordinate.
+#' @param niter How many to sample.
+#' @param seed input to set.seed (for repeatable results).
+#' @param coords_fix Coordinates to leave fixed. This can generate a different null distribution.
+#' @param ... Additional args passed to FUN.
+#'
+SampleUniformly = function( FUN, values, init, niter, coords_fix = NULL, seed = 0, ... ){
+  assertthat::assert_that( all( init %in% values ) )
+  objective = rep(NA, niter)
+  inputs = rep(NA, niter) %>% as.list
+  set.seed(seed)
+  for(ii in 1:niter){
+    x = sample(values, size = length(init), replace = )
+    x[coords_fix] = init[coords_fix]
+    inputs[[ii]] = x
+    objective[[ii]] = FUN(x)
+  }
+  return(list(inputs = inputs, objective = objective))
+}
+